VTO (voluntary time off) lets employees leave early when staffing exceeds workload. I initiated a 3+ month, field-led project to replace first-come, first-served allocation with explainable priority, operational safeguards and supervisor confirmation. I designed the business logic and used AI as a development partner, not as the decision maker.
Real shift pilot · Human-defined rules · Public demo uses synthetic records
LGLucas GiulianiMSc ISLI Global Supply Chain Management · KEDGE Business School
AT A GLANCE
Three months of work. One real operational problem.
An initiative developed alongside daily operations responsibilities, piloted on one shift and later adapted into an independent, synthetic portfolio edition.
AUTHOR-REPORTED3+ months
Analysis, design, iteration and operational project work
FIELD REPORT6+ weeks
Live pilot on a single operational shift
IN THE CODE7 criteria
Ordered individual ranking rules, plus coverage checks
PROJECT HANDOFF50+ pages
Illustrated guide for users and continuity
Reading the evidence: time and field figures are author-reported, not independently audited performance measurements. Code and handover figures describe documented project elements, not measured operational gains.
01 / THE PROBLEM
The old system rewarded speed, not fairness.
During my operations internship, I was managing a real shift while seeing how a simple allocation process could create tension, interrupt work and leave supervisors without a consistent explanation.
BEFORE / FIRST COME, FIRST SERVED
↗
A race for a place.
Employees left their work area to ask first. Previous departures did not influence priority. Decisions were difficult to explain consistently.
Observed issue: frustration, competing requests and avoidable distraction from the floor.
AFTER / A RULE-BASED APPROACH
✓
A reason behind each recommendation.
Volunteers are ranked on prior departure history, then checked against staffing and qualification requirements. A supervisor confirms the outcome.
Business objective: a fairer process that remains usable and safe for operations.
02 / THE BUILDING JOURNEY
The real challenge was making it work on the floor.
Building a screen was not enough. The tool had to fit existing management routines, technical boundaries and fragmented information, while I continued my day-to-day operational responsibilities.
THE CONSTRAINT
Improve the process without creating more work.
Supervisors needed a decision they could trust and explain. Operators needed a fair route to request VTO. The solution had to support both, without weakening shift coverage or relying on unrestricted access to internal systems.
3+months of iteration
Hundreds of hours of analysis, discussion, AI-assisted prototyping and adjustments informed by colleagues' feedback, alongside floor-management duties.
01 / INITIATE
See the problem. Obtain support.
I proposed the initiative and received approval from my direct Operations Manager to develop it and, with managers on my shift, to pilot it locally.
02 / DESIGN
Translate reality into business rules.
I analysed human constraints, priority disputes, staffing minima, critical skills and the site-specific constraints of a tightly controlled environment.
03 / BUILD & ADAPT
Connect the pieces without moving the data.
Information was spread across internal sources. I used several AI tools to challenge the approach and especially to help implement the interface, refining it repeatedly with colleague feedback.
04 / PILOT & HAND OFF
Prove usability before moving on.
After initial operator hesitation, the feedback I received by the end of my internship was consistently positive. A stable tool and illustrated documentation were handed over for continuity.
Evidence scope: the sequence, duration and feedback above are the project author's account. Approval to run a shift-level pilot is not the same as organisation-wide deployment or approval of this public portfolio.
03 / THE DECISION ENGINE
Business logic, made explainable.
The system recommends, it does not decide for a supervisor. Its individual ranking follows explicit, ordered business rules.
INDIVIDUAL PRIORITY
Seven rules. In order.
DETERMINISTIC
01Fewer departures this week
02Fewer departures over the last 56 days
03More waiting / compensation credits
04Fewer minutes of departure time this week
05Fewer minutes over the last 56 days
06Older date of last departure
07Deterministic tie-break
Additional group-selection logic applies to declared carpools. A high rank never bypasses operational eligibility checks.
▦
Protect the shift
Role minima, critical posts and qualified replacement checks are evaluated before a feasible recommendation.
◈
Respect exceptions
Onboarding restrictions, grouped carpools and eligibility rules add operational context beyond the fairness ranking.
✓
Keep a human accountable
Supervisors review the proposed allocation, record actual departures and provide reasons for exceptional overrides.
Explore the public portfolio edition with synthetic employees and in-memory changes. The actual decision workflow is ready to test, with no sign-in or setup.
LIVE PRODUCT PREVIEW · SYNTHETIC DATA
01
Prepare
Review the people, volunteers and available places.
02
Calculate
Apply history-based priority and staffing constraints.
03
Review
Inspect the recommendation and any qualified replacements.
04
Confirm
Record actual departures and update the history.
Start with a prepared scenario: 24 scheduled employees, 8 volunteers and 4 places. Changes are temporary; reloading restores the example.
05 / WHAT THE PILOT SHOWED · QUALITATIVE
From initial doubts to a tool people trusted.
The most meaningful change was not a dashboard metric. It was making departure requests less about who reached a supervisor first and more about a consistent, explainable process.
THE HUMAN SIDE
Early reluctance. Positive feedback by handoff.
At the start, some operators were reluctant to change a familiar process. I continued adjusting the tool around their feedback and the needs of managers. By the end of the internship, all of the feedback I personally received about the initiative was positive.
01StartInitial operator hesitation
→
02IterationFeedback-led adjustments
→
03HandoffPositive feedback received
VALUE FOR BOTH SIDES
Not just fairer for operators.
Operators no longer had to race to a supervisor to secure priority. Managers gained an explainable order of priority, operational coverage checks and a way to review exceptional decisions. These are process and product capabilities, not quantified productivity gains.
Human-reviewedManager confirms outcomesRule-basedNo AI selects employees
AUTHOR-REPORTED PILOT
The scope, without invented KPIs.
6+ weeks on one shift, approximately 30–40 allocation events managed. The pilot was locally approved and documented; all-site adoption has not been verified.
WHAT WAS NOT MEASURED
Improvement, not a claimed percentage.
Frustration appeared to decrease in day-to-day discussions, but there was no controlled satisfaction survey, audited baseline or statistically measured productivity gain. Positive feedback reflects the comments I received, not every employee's opinion.
06 / BUSINESS INTELLIGENCE · SYNTHETIC DATA
From recording departures to understanding patterns.
Added after the operational project, the analytics dashboard connects offers, employee interest and actual departures.
In this synthetic sample, NS has the highest participation rate. Each eligible employee is counted once per offer received.
03 / REQUEST OUTCOMES
Interest is not the final outcome.
938Resulted in departure
459No departure recorded
1,397 request instances, not unique people. Unfulfilled requests include different outcomes; the full dashboard explains them.
Source: the demo’s built-in synthetic event history. These charts describe the sample, not the professional pilot or measured business impact.
07 / AUTHORSHIP & AI TRANSPARENCY
I designed the decisions. AI helped build the software.
I worked with multiple AI systems over hundreds of hours of exchanges. They challenged my thinking and contributed substantially to the technical interface and implementation. They did not replace operational judgment or receive employee data.
MY ACCOUNTABILITY
Process ownership, not a claim to have hand-coded everything.
The hard work was defining the right problem, formalising decisions that would hold up on a real shift, choosing how to handle exceptions, and iterating until the tool made sense to both operators and supervisors.
Generative AI assisted development. The working allocation engine itself is deterministic and rules-based: human-specified priorities and safeguards produce a recommendation, and a human supervisor confirms departures. Spreadsheet structures, not their real contents, were shared with AI tools during development.
IMPLEMENTATION / LIMITS
Where did the original operational data live?
According to the author's account, operational information remained within the employer-controlled internal environment, with restricted network access. The initial application used a shared internal data file; the public portfolio edition does not connect to it.
How was the project tested and adopted?
The project was tested on an actual shift over six or more weeks, with feedback used to improve the tool. Documentation refers to automated checks, but the original test suite has not been independently reproduced in this public package. Positive feedback is qualitative and not an independently audited performance statistic.
Does a pilot imply a company-wide rollout?
No. The direct Operations Manager and managers on the shift supported a local pilot. The tool and documentation were handed over, but verified daily adoption across other shifts has not been established.
Is this public demo the original production system?
No. It is an independent, adapted portfolio edition using synthetic fixtures and an in-browser example. It does not use the internal shared data source. The later analytics section is also a portfolio extension.
08 / PRIVACY, SECURITY & GDPR
Real data stayed internal. The public demo is synthetic.
Privacy is not a slogan. There are three different data contexts here, and the claims we can make about each are deliberately separated.
01PROFESSIONAL PILOT / AUTHOR ACCOUNT
Employer-controlled environment
Real operational information stayed in the employer's internal systems and restricted shared environment. I did not send employee records or operational datasets to external AI services; AI tools were shown spreadsheet structures only.
A manager-authorised shift pilot does not establish a formal data-protection assessment or permission to publish employer materials.
02PUBLIC EDITION / CODE CHECKED
Fictional records, local interactions
The downloadable site bundle contains synthetic employee records and event histories, with no connection to the original employer data. Allocation calculations and example edits run in the browser. No app-level endpoint sending VTO records was found in the inspected files.
Some layout preferences may persist in your browser. Visitors should not import real people or confidential data into a public demonstration.
03WEBSITE DELIVERY / TO VERIFY
Hosting is not zero processing
Cloudflare necessarily receives web requests and may process IP addresses, security signals and other connection metadata. Hosting logs, network cookies, retention and geographic processing depend on the deployed Cloudflare configuration.
Those account-level settings have not been audited. This site does not claim a GDPR certification or blanket compliance.
A PROJECT YOU CAN ACTUALLY EXPLORE
Three months of problem-solving. One working decision tool.
Designed, piloted and documented by Lucas Giuliani MSc ISLI Global Supply Chain Management · KEDGE Business School